Agentic commerce: Enabling agents to understand, anticipate and act

What you should know:

• Agentic commerce could reduce the effort required to discover, evaluate and act on relevant commercial opportunities by combining personal context with user-directed assistance.
• Qualcomm Technologies and Mastercard demonstrated how on-device intelligence, XR glasses, a smartphone and agent-ready commerce services can work together while keeping user intent, consent and control central to the experience.
• The future of agentic commerce depends on agents and an ecosystem of partners that can move beyond recommendations to complete bounded actions on a user’s behalf.

For decades, digital commerce has followed a familiar pattern: users discover products and services, search for information, compare options, evaluate tradeoffs and then complete a purchase. Technology has made each step easier, but the overall cognitive burden still sits largely with the user. Agentic commerce has the potential to change that experience, but only if personal agents can work with commerce and payment services while keeping the user in control. That requires collaboration across the ecosystem, including the work Qualcomm Technologies and Mastercard demonstrated at Snapdragon Summit 2026.

AI agents are creating a new paradigm. They can help people understand choices, anticipate needs, evaluate opportunities and act. Yet as models and agents become more capable, intelligence itself may become increasingly abundant. One resource will not: user attention. Every notification, recommendation, offer and approval request competes for finite time and cognitive bandwidth. A useful agent should therefore optimize not for the number of interactions or transactions, but for the value created from each moment of attention. It should know what matters, whether now is the right time, when to ask, when to act and when to remain silent.

The agent must also be trustworthy. Users will only delegate decisions and actions to an agent if it consistently demonstrates good judgment and reliable execution. Without confidence in the agent's recommendations, transaction integrity and ability to recover from errors, any reduction in cognitive load will be outweighed by the need for users to continuously monitor and correct its behavior.

Even a highly capable agent needs access to products, experiences, merchants, payment capabilities and fulfillment. No single participant needs to own the full stack. Qualcomm Technologies can help enable  personalized on-device agent experiences using device-level AI and security features, while ecosystem partners can provide the specialized commerce and payment endpoints needed to complete real-world tasks.

How AI agents understand user intent and context

People operate across multiple timescales. They have immediate context, current goals, long-term interests, changing priorities, budget constraints, brand sensitivities and dependencies between decisions. Their needs also vary with location, activity, relationships and current state of mind.

A personal AI agent should combine these signals to understand not only who the user is, but where they’re headed and what follow-up action might be most useful to them. This understanding may draw from voice, vision, location, device interactions and user-approved personal knowledge. On-device agency is critical to keeping sensitive context local while enabling timely, low-latency experiences. When connected with commerce partners such as Mastercard, those on-device capabilities can extend beyond understanding and recommendation to support real-world discovery and transaction experiences.

Qualcomm Technologies and Mastercard demonstrate agentic commerce in action

At Snapdragon Summit 2026, our joint demonstration with Mastercard illustrated how personal AI agents can move beyond understanding and recommendation to support discovery and commerce experiences. A user wearing XR glasses connected to a smartphone enters a hotel lobby and encounters a range of attractions and experiences. The glasses provide a  hands-free interface for observing the environment, while the phone supports the personal agent and broader workflow.

In the current experience, the user deliberately signals interest through voice or a button press. This choice prevents the agent from continuously surfacing offers and visibly reinforces user control. One potential future direction is a more proactive experience: as the agent learns the user’s interests, trajectory and context, it can selectively recognize high-value moments and guide the user toward opportunities likely to matter, within boundaries the user controls. This selective approach is intended to prioritize opportunities based on relevance rather than availability alone.

The agent is designed to combine contextual signals with user preferences to identify commercial opportunities that may be relevant. Mastercard’s Priceless capabilities help connect that personal understanding to available commercial experiences. The user reviews the relevant option and supplies consent, intent and any additional purchase constraints, ensuring that agency remains user-directed rather than autonomous. The on-device agent then orchestrates the remaining workflow, while Mastercard’s agent-ready commerce and payment capabilities, using the PayOS framework, support transaction execution. The agent returns purchase confirmation to the glasses or phone.

The Qualcomm Technologies and Mastercard demonstration showcases an end-to-end agentic commerce workflow: learning about the user, identifying relevant commercial opportunities, capturing user intent and completing a transaction through agent-enabled commerce services on user devices.

How AI agents move from recommendation to transactions

Traditional digital experiences wait for a query. Proactive agents can recognize when context and likely intent create an opportunity to help before the user begins a search.

The agent must decide whether an opportunity is sufficiently relevant to surface, whether it should gently guide the user toward clarifying intent and whether the potential benefit justifies consuming attention. User-defined boundaries should govern how proactive the experience may be. The long-term objective is bounded proactivity: anticipate intelligently, engage judiciously and preserve user control.

Commerce requires the agent to determine what is purchasable, compare options, account for constraints and recognize when the user must be involved. Once clear intent and consent are established, the agent is built to take over the routine steps that create friction, including navigating the transaction flow and returning a confirmation.

This is where agentic commerce differs from a conventional shopping experience. The user should not have to repeatedly restate preferences, move between applications or manually execute every step. The agent carries context forward and performs bounded actions on the user’s behalf. They may also capture the user’s intent to purchase and execute the operation at a more suitable time (e.g., wait until price drops below a given threshold). Trust, user consent, credential protection and reliable execution will be essential as agentic commerce evolves, helping enable users to confidently delegate more sophisticated tasks to their agents.

The future of agentic commerce: maximize value, not engagement

Today’s commerce is largely search-driven. Tomorrow’s will be increasingly intent-driven and agent-mediated. The interface may be a phone, glasses, a PC, a vehicle or any another connected device, but the underlying challenge will be the same: understand what the user is trying to accomplish, decide whether attention is warranted and safely perform the work that follows.

The best agent will not be the one that speaks most often, shows the most offers or maximizes the number of transactions. It will be the one that earns trust by making the right intervention at the right moment, then removes friction through secure and bounded action.

Agentic commerce shows how personal understanding, proactive intelligence, trusted execution and ecosystem connectivity can help agents move from answering questions to delivering outcomes, while treating the user’s attention as the scarce resource it is. The Qualcomm Technologies and Mastercard demonstration illustrates why this future will take an ecosystem: on-device intelligence can interpret user context and orchestrate workflows, while commerce partners provide the specialized services needed to translate that understanding into real-world transactions and experiences.

Go Deeper

What is agentic AI, and how is it different from a traditional AI assistant?

Agentic AI moves beyond answering individual questions to understanding goals, carrying context forward and helping execute a sequence of actions. In this blog post, that shift is reflected in the movement from commercial recommendations to user-directed transaction execution. 

Why is on-device and cross-device intelligence important for personal AI agents?

Personal agents need access to relevant context while remaining responsive as users move between devices and environments. This blog post shows that model through XR glasses that perceive the environment and a connected smartphone that supports the personal agent and broader workflow. 

How can personal AI become proactive without taking control away from the user?

This blog post describes bounded proactivity: the agent assesses potential relevance, engages selectively and involves the user when intent, consent or additional constraints are required. In the Mastercard demonstration, the user deliberately signals interest, reviews the proposed option and provides consent before the workflow proceeds.